def main():
args = parse_args()
metrics = []
for func in args.selected:
try:
r = func()
if isinstance(r, list) and r and isinstance(r[0], MetricData):
metrics.extend([result.metric() for result in r])
elif isinstance(r, MetricData):
metrics.append(r.metric())
except: # noqa
t, v, tb = sys.exc_info()
backtrace = ' '.join(traceback.format_exception(t, v, tb))
r = MetricData.single('cephlm.probe.failure',
Severity.fail,
'{error} failed with: {check}',
dimensions={
'component': 'cephlm-probe',
'service': 'ceph-storage'
},
msgkeys={
'check': func.__module__,
'error': backtrace.replace('\n', ' ')
})
metrics.append(r.metric())
# There is no point in reporting multiple measurements of
# cephlm.probe.failure metric in same cycle.
check_failures_found = []
for metric in metrics:
if metric.get('metric') == 'cephlm.probe.failure':
check_failures_found.append(metric)
if check_failures_found:
# Remove all except one instance
for metric in check_failures_found[:-1]:
metrics.remove(metric)
else:
r = MetricData.single('cephlm.probe.failure',
Severity.ok,
'ok',
dimensions={
'component': 'cephlm-probe',
'service': 'ceph-storage'
})
metrics.append(r.metric())
FORMATS[args.format](metrics, args.pretty)
def osd_stats():
"""
Publishes the osd statistics
"""
metric_dict = {
'up': OSD._up_count,
'up_out': OSD._up_out_count,
'down': OSD._down_count,
'down_in': OSD._down_in_count,
'total': OSD._total_count
}
INVALID_VALUE = -1
result = list()
probe_failed = False
try:
osd_stats = OSD._stats()
except (exc.CephLMException, exc.CephCommandException,
exc.CephCommandTimeoutException) as e:
probe_failed = True
for metric_state, func in metric_dict.iteritems():
name = "cephlm.osd.%s_count" % metric_state
if probe_failed:
value = INVALID_VALUE
msg = "Probe error: Command 'ceph osd tree' failed"
else:
value, msg = func(osd_stats)
msg = "OSD(s) %s" % msg if msg else "No OSD(s)"
msg += " is/are in cluster" if metric_state == 'total' \
else " is/are %s" % metric_state
base_result = MetricData.single(name, value, message=msg)
result.append(base_result)
return result
def pool_stats():
"""
Publishes the pool statistics
"""
result = list()
INVALID_VALUE = -1
probe_failed = False
metric_dict = {
'count': 'count',
'total_objects': 'objects',
'usage_bytes': 'size_bytes',
'top_three_by_usage_bytes': 'top_pools_by_size',
'top_three_by_objects': 'top_pools_by_objects',
}
try:
pool_dict = Pool._stats()
except (exc.CephLMException, exc.CephCommandException,
exc.CephCommandTimeoutException) as e:
probe_failed = True
msg = str(e)
for metric_name, state in metric_dict.iteritems():
name = "cephlm.pool.%s" % metric_name
if probe_failed:
value = INVALID_VALUE
elif 'top_three' in metric_name:
msg, value = Pool._pools_by_metric(pool_dict, state)
else:
msg, value = Pool._return_total_metrics(pool_dict, state)
base_result = MetricData.single(name, value, message=msg)
result.append(base_result)
return result
def pg_stats():
"""
Function to aggregate all metrics
"""
msg = ''
INVALID_VALUE = -1
probe_failed = False
try:
pg_stats = PG._stats()
except (exc.CephLMException, exc.CephCommandException,
exc.CephCommandTimeoutException) as e:
probe_failed = True
msg = 'Probe error: ' + str(e)
if probe_failed:
value = INVALID_VALUE
else:
value = pg_stats.pop('count')
for pg_state, count in pg_stats.iteritems():
msg += '%s=%s, ' % (pg_state, count)
msg = msg[:-2]
name = "cephlm.pg.count"
base_result = MetricData.single(name, value, message=msg)
return base_result
def capacity_stats():
"""
Publishes the capacity statistics
"""
metric_list = ['total_bytes', 'used_bytes',
'available_bytes', 'perc_utilization']
msg = ''
result = list()
capacity_dict = dict()
INVALID_VALUE = -1
probe_failed = False
try:
capacity_dict = Capacity._stats()
except (exc.CephLMException, exc.CephCommandException,
exc.CephCommandTimeoutException) as e:
probe_failed = True
msg = str(e)
for metric_name in metric_list:
name = "cephlm.capacity.%s" % metric_name
value = capacity_dict[metric_name] \
if not probe_failed else INVALID_VALUE
base_result = MetricData.single(name, value, message=msg)
result.append(base_result)
return result
